rgem227/knoweldge-base

This repository contains the phase 1 minimum viable file knowledge base.

• Backend: FastAPI
• Database: SQLite
• File storage: local filesystem
• Task queue: placeholder for later phases

PostgreSQL and object storage are reserved for a later migration. The current version does not require Docker, MinIO, or a running database service.

1. Copy the root environment template to backend/.env:

Copy-Item ..\.env.example .env

1. Install dependencies:

cd backend
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

1. Initialize the SQLite database:

python -m app.db.init_db

1. Run backend:

uvicorn app.main:app --reload

From the backend directory:

set PYTHONPATH=.
python -m pytest

PowerShell:

$env:PYTHONPATH = "."
python -m pytest

Initialize or recreate the SQLite tables:

python -m app.db.init_db

Runtime files are created under backend/data/:

• knowledge_base.db: SQLite database
• storage/: uploaded files and generated Markdown files

The existing Docker Compose file is reserved for the later PostgreSQL and object storage migration. It is not required for the current phase.

Service start, restart, stop, and status commands are provided by the separate PowerShell control files under scripts/.

• Administrator-created users, login, and JWT authentication
• Public knowledge bases
• Nested folders with breadcrumb navigation
• Folder-scoped document listing and upload
• Owner-only folder rename and empty-folder deletion API
• Single-file and batch upload
• Upload support for PDF, text/Markdown/HTML, images, PowerPoint, Word, and Excel files
• Document metadata and tags
• Original file preview and download
• In-app reader for text, Markdown, PDF, DOCX, XLSX, PPTX, and images
• Browser-local preview for PDF, DOCX, XLS/XLSX, and Markdown using Vue-based preview components, with PPTX rendered by the bundled pptx-renderer
• Best-effort text extraction for legacy DOC, XLS, and PPT files
• Owner-only document editing
• Owner-only soft delete and restore
• Recycle bin listing
• SQLite database initialization
• Local filesystem storage with path traversal protection
• Administrator-only user management and audit log pages
• Per-user MCP Token expiration policies, including persistent long-lived Tokens
• In-app upload progress, safe HTML preview, document owner/uploader metadata, and administrator MCP invocation logs
• Cross-scope copy for documents and folders, including recursive folder copies and physical local-storage duplication

Operational configuration is controlled through .env.example, backend/app/core/config.py, and the scripts under scripts/. The repository also includes a read-only REST API, OpenAPI endpoints, MCP integration, and the knowledge-base Skill under skills/knowledge-base/.

Log in and receive a token:

$body = @{
  username = "demo"
  password = "password-123"
} | ConvertTo-Json

Invoke-RestMethod `
  -Method Post `
  -Uri http://127.0.0.1:8000/api/v1/auth/login `
  -ContentType "application/json" `
  -Body $body

Upload a document:

$token = "<access_token>"
$knowledgeBaseId = 1

curl.exe `
  -X POST `
  "http://127.0.0.1:8000/api/v1/documents/upload" `
  -H "Authorization: Bearer $token" `
  -F "knowledge_base_id=$knowledgeBaseId" `
  -F "tags=制度,测试" `
  -F "file=@D:\path\to\guide.txt"

The generated API documentation is available at /docs.

The frontend uses vue3-office-preview for DOCX, PPTX, XLS, and XLSX, @vue3-office/vue-pdf for PDF, and @deot/docs-markdown for Markdown. Files are fetched through the authenticated API into browser memory and rendered locally. No LibreOffice, ONLYOFFICE, Docker, or Python SDK is required for this preview path.

The original file remains in local storage. Markdown extraction and the legacy .doc, .xls, and .ppt fallback remain handled by the backend.

Create a folder under a knowledge base:

$body = @{
  knowledge_base_id = 1
  name = "实验方案"
} | ConvertTo-Json

Invoke-RestMethod `
  -Method Post `
  -Uri http://127.0.0.1:8000/api/v1/folders `
  -Headers @{ Authorization = "Bearer $token" } `
  -ContentType "application/json" `
  -Body $body

Upload into a folder by adding folder_id to the multipart form. The web page does this automatically for the currently selected folder.

The knowledge base selector defaults to 全部. In this mode the page shows folders and root documents across all public knowledge bases. New folders are assigned to the knowledge base selected in the dialog; uploads from 全部 use the first available knowledge base unless a folder is selected.

The project includes a separate MCP service for agents. It shares the same SQLite database, local filesystem storage, JWT secret, and document reader as the FastAPI application. It supports local stdio and remote Streamable HTTP transports. Phase 1 exposes read-only tools:

• list_knowledge_bases
• list_documents
• search_knowledge
• get_document
• get_document_metadata

The MCP process does not expose upload, move, delete, or folder-management operations. It uses the JWT identity from KB_MCP_TOKEN, so an agent cannot see documents that the corresponding user cannot read.

For local stdio clients, KB_MCP_TOKEN can be omitted. The agent must then call the authenticate tool once with the user's username and password; the short-lived session remains only in the MCP process memory.

Create the MCP environment once from PowerShell:

cd backend
python -m venv .mcp-venv
.\.mcp-venv\Scripts\python.exe -m pip install -r requirements-mcp.txt

Set the same database and storage settings as the API, then put a login access_token in KB_MCP_TOKEN:

$env:PYTHONPATH = "."
$env:KB_MCP_TOKEN = "<access_token>"
.\.mcp-venv\Scripts\python.exe -m app.mcp_server

The process speaks MCP over stdin/stdout. Do not add ordinary print() output to the service; diagnostics must go to stderr so the protocol remains valid.

For remote Streamable HTTP mode:

$env:PYTHONPATH = "."
$env:MCP_TRANSPORT = "streamable-http"
$env:MCP_HOST = "0.0.0.0"
$env:MCP_PORT = "8020"
$env:MCP_PATH = "/mcp"
$env:MCP_STATELESS_HTTP = "true"
.\.mcp-venv\Scripts\python.exe -m app.mcp_server

The remote endpoint is https://your-domain/mcp after HTTPS reverse proxy configuration. Remote clients must send Authorization: Bearer <access_token>. Configure the remote client with the same endpoint and protected Bearer credential used by the API connection.

The browser page /mcp-token.html reads the current logged-in account and requests an MCP Token without reading or storing the account password. The administrator page /admin.html is protected by the administrator role and allows user status, role, password, and MCP Token expiration policy changes.

Run the local protocol smoke test:

$env:PYTHONPATH = "."
.\.mcp-venv\Scripts\python.exe scripts\mcp_stdio_smoke.py